Select text to annotate, Click play in YouTube to begin
00:00:08
thank you well thank you so much and I'm delighted to be here thank you for putting on such an amazing symposium and uh you're giving me the opportunity to share some thoughts with you in in long form
00:00:24
so uh what I'm going to talk about today is basically uh the fields of collective intelligence in unconventional spaces and if anybody would like to get in touch with me afterwards all of the
00:00:36
primary papers the software everything else is is here so I want to start by uh just showing one of my one of my heroes Alan Turing we all know that he was a forefather of artificial intelligence he
00:00:48
was very interested in uh this idea of of creating uh cognitive systems and various kinds of intelligence um and uh this is this is this is well known what's maybe less well known is that the same person was really
00:01:02
interested in morphogenesis he wrote one of the very early uh papers on uh on trying to understand self-organizing uh chemical systems and you might wonder why it is that the same person would be interested in intelligence and
00:01:14
morphogenesis and I think that uh he saw a very deep parallel between these two these two areas of study and I think he was right on the money in the sense that I think these are actually very much the same question
00:01:27
now most of the time when we think of intelligence we think of something like this so here's a brain and we think of okay here's a here's a what quote unquote centralized intelligence of some some animal um and then we look at things like this
00:01:39
ant colonies and so on and sometimes we talk about them as uh distributed or collective intelligence this warm cognition and so on you know Ricard Soleil calls them liquid brains but actually it's it's important to realize that this is just a matter of scale
00:01:52
because if one Zooms in you find out that we are all in fact Collective intelligences and we are all made of something like this so this is a single cell this is a lacrum area it has no brain it has no nervous system it has no
00:02:05
cell to cell uh Communications or stem cells or any of the things that we're used to having in bodies and what it what it does is it handles all of its local needs so it's a morphological uh behavioral metabolic and so on at the
00:02:18
Single Cell level very competently and the amazing thing about this is that we all have taken the same uh Journey from um what what might be called just
00:02:33
physics which is an oocyte right a a set of uh a set of chemical reactions that people would look at and would think okay this this says no cognitive capacities is just just in physics and chemistry and yet uh slowly so over a
00:02:46
matter of some months and and maybe years this process gives rise to these amazing morphologies and in fact to high-level human cognition that has second order you know metacognition and self-awareness and so on and so this is
00:02:59
a journey across the Cartesian cut that we all take if you follow your own uh your own development back far enough you will find this and before that you will find just so it's quote unquote just physics so it's really interesting to to
00:03:11
ask how how this transition works this this smooth transition and so the main points that I'm going to try to transmit today are these first that biology uses a kind of multi-scale Competency architecture of nested problem solvers
00:03:24
and that navigation is a really Central concept to try to understand navigation of spaces in particular is a central concept to try to understand this I'm going to claim that goal directedness is a is a critical and variant for
00:03:36
recognizing building and uh relating to various unconventional agents and I'll describe a kind of cognitive boundary model for the scaling of goals I'm going to spend about half the talk showing you
00:03:49
one specific example in detail which is this idea that pattern formation biological pattern formation is literally the behavior of a collective intelligence of cells in a space known as morphospace and I'm going to show you
00:04:03
that bioelectrical networks are the protocognitive medium the ancestor of brain function and this this idea actually has some very uh practical impacts on biomedicine this is not just
00:04:15
the sort of philosophy this has very specific applications and in the end I'm going to show you how synthetic bioengineering provides a really astronomically large option space for new bodies and new minds that don't have
00:04:28
standard evolutionary backstories so we'll get to that towards the end so single cells have some really uh interesting spatial competencies that we can start off by thinking about so this is a diatom this is a single cell in
00:04:41
that particular very particular structure is something that it can reliably acquire this is a collection of uh several acetabularia algae each one of these things is in fact a single cell it has one nucleus the whole thing is
00:04:54
maybe uh you know six or seven centimeters long it has some Roots it has a stock it has a cap here the whole thing is just one cell and so this is quite interesting to ask how it is that a single cell can have such such profound
00:05:07
morphological differentiation but these uh these competencies are not only spatial they are also behavioral and so what you're seeing here is a slime mold this is a fisarin polycephalum it is the
00:05:19
whole thing is one cell it could be you can grow to be very large but the whole thing is one cell and so when you place it in the middle of this this petri dish the little white circles are glass discs there's nothing there's no food there's no chemicals they're completely inert
00:05:30
glass so what it will do as it grows during the first few hours it sort of grows evenly outward like this and what it's doing during this time is sending out vibrations into the medium it pulses and sends vibrations into the medium
00:05:44
reading back the vibrations that that return so it's almost a kind of sonar it's sensing uh sensing the properties of the medium and by doing this it can actually build a map of its environment and then reliably make its way over to
00:05:57
the heavier Mass so it can reliably detect three disks over one and you can do all sorts of fun experiments of stacking them one on top of each other and distributing them but it has an amazing ability to sense the mass in its vicinity and to make both morphogenetic
00:06:09
and behavioral decisions based on that information okay so so you can you can see this this cell navigating in this in this space according to the to the biomechanical information that it has so I'm going to show you four basic things
00:06:22
today so so first I'm going to um try to uh introduce some some biology that you may or may not have seen before designed to really kind of stretch our idea of what is an organism what is a mind what
00:06:35
does it mean to navigate space and so on and then then we'll get into this very specific examples of bioelectricity and morphogenesis and then uh we'll talk about some of these uh novel organisms so uh the kind of the traditional view
00:06:48
is is of these um discrete static natural kinds you have a particular animal maybe you have a rat that you're studying in a maze or a human subject or whatever you have and that's you know that's kind of that's kind of your unit but I want to I want to go beyond this
00:07:01
this sort of Garden of Eden View and really emphasize that agents that do these intelligent things are incredibly plastic so here's one familiar example here we start with a caterpillar this is
00:07:12
a creature that lives in a in a largely two-dimensional world that crawls around on flat surfaces it chews leaves and it has a brain appropriate for that purpose it has to turn into a butterfly which is going to live in a very much a
00:07:25
three-dimensional World it has to fly it has to drink nectar uh and uh and it needs a completely different brain this is a soft-bodied robot this is a hard-bodied you know needs a controller suitable for hard kinds of bodies and so
00:07:37
in between what happens is during this process the brain is largely largely destroyed most of the connections are broken down most of the cells die and and the new brain is rebuilt during the
00:07:49
lifetime of the organism so you know this this this kind of change makes the the confusion of puberty seemed like um real Child's Play you know in the sense that this is a single agent radically changing it's it's its brain and its
00:08:02
body and the amazing thing about it is that uh memories that the caterpillar acquires are retained in the moth or butterfly so this has been shown so despite the despite the uh disaggregation of the brain actually
00:08:15
some information is able to make it across in fact this can be even more radical these are these are planarium these are flatworms and you'll see a lot more about them in this talk the planaria regenerate any part of their body so you cut them into pieces of the
00:08:27
record is something like 275 pieces each piece regrows exactly what's missing including the brain and you get perfect little worms so um James McConnell back in the 60s made an observation and uh it
00:08:39
was um uh you know quite controversial at the time but we repeated it and and actually discovered he was he was absolutely right what happens is that if you if you train these planaria to recognize a particular region of their environment let's say one with these
00:08:52
little little bumps laser etched into it as the place where they get fed you can then amputate The Head and the Brain I mean these animals have a true centralized brain amputate The Head and the Brain leave the tail the tail sits there for eight or nine days doing
00:09:05
pretty much nothing eventually it grows a new fresh brain and somehow that information is imprinted onto this new brain and these animals show show retention of that of that memory so the
00:09:16
information moving through the body being imprinted from from one tissue onto another is something that uh we really need to start to understand and in fact uh Beyond uh Beyond those kind of experiments you can do this
00:09:29
invertebrates we did this in the tadpole so this is a tadpole of the Frog xenobus lavis um here's the brain here are the nostrils here's the mouth the gut and the tail one thing you'll notice is that what we've done is we've prevented the
00:09:40
normal eyes from forming we've put an ectopic eye on the tail so this is a this is an eye that forms on the tail and the amazing thing about these animals is that with no long in fact with no period of evolutionary
00:09:52
adaptation to this new configuration they can see perfectly well out of these eyes so we've made a device that tests them on visual training cues and sort of automates the same thing we use to test the planarian memory it's an automated
00:10:03
Behavior testing device and what we found is that this eye finds itself in a completely novel environment surrounding my muscle instead of near the brain puts out an optic nerve that optic nerve might connect to the spinal cord it does not go up to the brain the brain that
00:10:17
evolved for tracking visual data from these two locations suddenly gets information from some weird itchy patch of tissue on its tail no problem it recognizes that as visual data and these animals behave quite well in in learning
00:10:29
assays so so navigating um navigating your world with a radically different configuration of sensory and and processing organs so what we really see
00:10:42
in biology is that not only are we nested dolls structurally I mean we all everyone knows that groups consist of individuals made of organs tissues cells and so on but actually each one of these levels has competency it solves problems
00:10:56
in its own space and there are lots of different kinds of spaces and that multi-scale architecture which is which is kind of unique it's something that um in engineering we're still really not
00:11:08
not able to uh re recapitulate although we're getting there uh shows competency in many many different spaces now now the one thing I'd like to I'd like to do fundamentally is that uh generalize this idea of being able to
00:11:21
perform intelligently in some space so typically typically we think about three-dimensional spaces so uh Behavior moving the body in three-dimensional space so it's very easy for us to to
00:11:35
recognize intelligence in those spaces because all of our sense organs Point outward and from the time you were very little you were collecting data on object medium-sized objects moving at medium speeds around you in three-dimensional space and and we know
00:11:48
how to recognize you know birds and and mammals and other things and be behaving intelligent what we don't have are senses that directly show us for example imagine that you had a you were born with a sense that where you could
00:12:00
actually feel your blood chemistry all the different things going on in your blood physiology and all of the things that your liver pancreas and other organs were doing in that case we would have a training set that would allow us to understand intelligent navigation of
00:12:13
other spaces so that would be physiological space so the space of physiological parameters um it might be a transcriptional space so the space of gene expression all the different gene expression domains or in
00:12:25
fact morphospace this the space of all possible anatomical configurations we'll talk more about that I want to show you um and and this this the the the fact that we're not uh familiar with these other spaces means that when we make uh
00:12:39
claims about the cognitive level or the intelligence level of other systems we're really taking an IQ test ourselves what we're really saying is this is what we've recognized the system to be doing but we can't really assume that uh that we're smart enough to know exactly what
00:12:51
it's doing and what it's solving in all of the different spaces I want to show you a simple example again these are plenty these are the planaria and what we found is that if we put planaria into a a solution of barium chloride so
00:13:03
barium is a non-specific potassium channel blocker it prevents these cells from exchanging potassium with the outside world so when you do this literally their heads explode okay their heads just overnight they blow up but the
00:13:16
amazing thing is that if you keep the rest of the worm in barium over the next couple of weeks they regenerate they grow a brand new head and the new head does not care about barium whatsoever the new head is barium adapted now
00:13:27
that's kind of amazing and so we asked a simple question we just looked at the transcriptomes of naive worms versus barium adapted heads and we just asked well what gene expression is different okay what's different about these barium
00:13:39
adapted heads and one thing we found is that there's only a handful of genes there's less than a dozen genes that are in fact different and so this this system is able to figure out which genes to up and down regulate so that it can
00:13:51
do its business without being able to pass potassium properly but the incredible thing about this is that barium is not anything that planaria ever see in the wild this is completely uh a completely unnatural novel stressor
00:14:03
so so it's implausible to think that at some point there was evolutionary pressure to develop a response to Barium so what you're really talking about here is you're in the you're in the space of gene expression which is you know let's
00:14:15
say if they have 20 000 G's it's a massively um you know a very high dimensional space and you need to walk in that space to find the the exact uh genes that are going to solve your physiological stressor now you don't
00:14:28
have time to sort of randomly flip jeans on and off first of all because it'll it'll most likely kill you before you find the good concentrate the good combinations also also there's no there's no time for that these cells don't turn over that fast it's not like
00:14:41
a bacterial evolutionary system so it's still very much an open question how how do they navigate this this transcriptional space to to solve these kinds of solve these kinds of problems so this this to me is is very much an
00:14:55
example of of intelligence in the sense of problem solving taking what you already might know maybe there's information about what to do in a case of epileptic is seizures or so on and bringing it to a new a new scenario and so so I've been working on this on this
00:15:08
framework now it's called t-a-m-e tame for it stands for technological approach to mind everywhere and the goal of this framework is to be able to uh recognize uh create and relate to truly diverse
00:15:21
intelligences so we need to handle of course the things we're familiar with the birds uh primates and so on but also weird Colonial organisms and swarms and things like that um of course all the new things that are
00:15:33
being produced by synthetic morphology and synthetic biology approaches and possible exobiological agents because studying uh just the natural systems that that are here on Earth and
00:15:45
making conclusions about biology from them is a little bit like testing your your theory on the same data set that that created it right it's all just an N of one uh pass through um through a phylogenetic space so so I I tend to
00:15:58
think about all kinds of unconventional uh different different types of embodiments whether they've evolved design whatever and really think about this kind of this is a scale by rosenbluth wiener and Bigelow looking at all the different kinds of behaviors
00:16:10
that one might have in one space notice that it says nothing about what it's made of it doesn't doesn't say anything about having brains or being a specific kind of organism or scale temporally or whatever it's it's all it's all very functional so this is this is this is
00:16:23
how we think about things in my group and the key to this framework is that it has to move experimental work forward it has to enable new capabilities and um I'm going to show you I'm going to show you how that happens but first let's
00:16:34
just think about uh this this one particular kind of intelligence which I think is is super interesting um anatomical control as a as a collective intelligence first of all notice the basic thing that needs explaining which
00:16:47
is that we all Start Life as a collection of embryonic blastomeres and these this this is a cross section through a human torso so this is what uh what each of us has inside now look at all the in incredibly complex invariant
00:17:00
order all the all the tissues and structures everything is in the right place next to each other the right orientation size and so on so the first question is where is the shape encoded how do how how do these cells know to make exactly
00:17:12
this and you might be tempted to say DNA and genomes but we can read genomes now and what we see is that genomes directly code for protein structure so the genome specifies the micro level Hardware
00:17:24
that's present in every cell but there's nothing directly that you can read out in the genome about the Symmetry type of the organism the size the shape how regenerative it's going to be um this is very much an open problem of how cells know what to what to make and
00:17:37
when to stop uh as workers in regenerative medicine if parts of this are missing we'd like to know how to signal the cells to rebuild to do it again and as Engineers we'd like to know well what else is possible Right given
00:17:49
the exact same genome what else can we ask these cells to do or is this the only thing they could possibly do and you can you can sort of visualize forward the end game of this whole field is something like this it would be an anatomical compiler where you should be
00:18:02
able to sit down and draw at the level of anatomy the animal or plant that you want okay not not the not the pathways not genes but but the actual anatomy and uh if we knew what we were doing we would have a system that
00:18:15
um that compiled that description into a set of stimuli that would have to be given to cells to get them to build this particular thing in this case a nice three-headed planarian now we don't have anything remotely like this this is a
00:18:27
very long a long-term goal and the reason that it's really important is because if you think about it um pretty much every problem of biomedicine with the exception of infectious disease so birth defects uh
00:18:39
traumatic injury cancer aging degenerative disease all of these things boil down to one problem how do you convince cells to build the exact structure that you want if we solve that problem all of this goes away we'd be able to fix birth defects regenerate
00:18:51
limbs reprogram tumors all of this would would go away but it's a major major problem why do we not have an anatomical compiler yet so I want to be clear that despite the incredible progress in
00:19:04
genetics and molecular biology we Face very fundamental questions that have to do with not not with the molecular mechanisms but with the decision making so here's a simple example um here is this is a baby Axolotl so
00:19:16
axolotls are Mexican salamander there's a baby Axolotl and baby axolotls have legs there's a tadpole not the frogs and a slavis tadpoles do not have legs so now in my group we make something called the Frog level so this is a half it's a
00:19:30
it's a chimera half Axolotl half frog you can mix the cells they they cooperate with each other just fine they make something a frog a lot now I ask a simple question you've got the genome to the Axolotl you've got the genome to the
00:19:42
Frog how come we don't have any models that tell us where the Frog bottles will have legs they have no idea from from that information whether frog lots are going to have legs or not if they do have legs whether those legs will be made of Axolotl cells or also include
00:19:56
frog cells we we have no idea even though you've got the genetic information so it's really important to start to understand the algorithms because where where biomedicine is right now is that we're very good at manipulating molecules and cells and
00:20:09
getting information like this which which Gene and protein talk to which other Gene and protein we are a long way away from actually controlling a large scale Form and Function and in fact you can think about the kind of parallel
00:20:22
Journey that that computer science took this is what programming looked like in the 40s and 50s where in order to control the system you have to physically rewire it right you were pulling wires in and out you have to rewire it nowadays days uh for a joke I
00:20:35
say to all my students you know you're on your laptop you're going to go from from Microsoft Word to photoshop I want you to get out your soldering iron and start rewire you know start rewiring and of course they all laugh because because nowadays we don't need to do that we
00:20:46
understand that if your Hardware is good enough it's reprogrammable with stimuli with inputs with experiences not rewiring but of course modern molecular medicine is all about the hardware we're all we're all very excited about genomic
00:20:58
editing and protein Pathways and single molecule approaches and so I think the reason that we are still roughly where computer science was in the in the 50s is because we are we've been neglecting one important aspect
00:21:11
and that is multi-scale intelligence in biology now what do I mean by intelligence I mean uh what William James meant which is the ability to reach the same goal by different means and when uh when we talk about a pattern
00:21:25
and Anatomy when I talk about goals I mean regions of morphe space now what's morphe space morphe space is the uh the space of all possible configurations of some particular structure so if you're looking at snail shells for example
00:21:38
there are three parameters that you can Define and uh every shell is every every possible snail shell is some point within this amorphous space okay and this is this is an idea that's very that's very old in fact Darcy Thompson
00:21:51
in the 40s had this interesting example in his in his book on growth and form where he noticed that if you just deform certain animal shapes placed on a grid you apply specific deformations to the grid what you get are other other
00:22:04
species that do exist now at the time there was no molecular mechanism of course known there was not nothing really you know really known about this but I think I think now we can we can we can do some some very interesting things with this idea
00:22:17
so uh navigating these spaces changing your body shape to move from one region to another is not uh trivial because there may be local Minima there may be barriers there may be all kinds of things so that that's the task that we
00:22:30
face as as uh as morphogenetic agents now embryogenesis is is very good at this they're they're it's extremely reliable so you start off in uh as as this you know as this kind of pattern
00:22:42
and eventually you end up here and that's that's generally very low very low air um but we can already see actually that that this process is not simply a kind of pre-programmed hardwired walk in
00:22:54
morphospace because we can deviate in the in the middle so for example we can take this nice uh human embryo divided in half literally cut it in half and what you get are not two half organisms you get two perfectly normal monozygotic
00:23:07
twins this is where twins come from and so this is a regenerative event where each saw each half of this embryo basically realizes that it's actually not where it's supposed to be in morphe space um and uh and it needs to regenerate the
00:23:20
other half in order to make the correct changes to get to its goal and where it needs to go so this is not just embryonic uh for example back to this this um the salamander this um this Axolotl
00:23:32
um these guys regenerate their limbs their eyes their jaws their spinal cords portions of their heart and brain their ovaries they're incredibly regenerative as adults and what happens is that you
00:23:44
can amputate uh for example this limb at different positions no matter where you cut the cells will very quickly grow they will grow exactly what's needed to make a normal axolotlim and then they stop so this is a kind of example of anatomical homeostasis they will
00:23:57
continue working from wherever starting position until they get to the end until they get to where they're going parenthetically this is not just for for um for frogs and for you know acid levels and worms the human liver of
00:24:09
course is regenerative even the ancient Greeks knew that I have no idea how they knew that but but it seems like they did um dear every year regenerate massive amounts of of bone up to a centimeter and a half of new bone per day bone
00:24:22
vasculature innervation skin and even human children below a certain age can regenerate their fingertips if you just leave it alone it'll basically regrow cosmetically a very very nice outcome so we have we have some some ability to uh
00:24:35
to improve our our position in morphe space one of the most amazing things about it is that as a as a body as a living creature you can't count on your environment you can't count on in other
00:24:48
words you can't count on the environment being the same as it was before you can't count on not being perturbed during the developmental process maybe physiologically maybe metabolically maybe um with a parasite or a teratogen so you can't count on that and you still need
00:25:01
to get your job done in fact you can't even count on your own Parts being what you expected them to be now uh well of course we don't have any machines that can do this that can they can repair themselves after damage or put themselves together with diverse Parts
00:25:14
here this is this is one of my uh favorite examples of all time this is uh a cross-section through a newt um kidney tubule so so here's the Lumen of the tubule needs that these are the cells that make up the tubule normally
00:25:26
it's a let's say around eight cells that work together to form this kind of tubule one thing you can do is you can you can make uh you can make these cells what's called polyploid which means they have extra genetic material amazing
00:25:40
thing number one with the excess genetic material you still get a perfectly normal Newton no problem that they're all kinds of extra um information around them no no no no problem but what does happen is that the cells get much bigger and as the cells
00:25:53
get physically bigger fewer of them are needed to make the exact same shape Lumen now that's that that's amazing thing number two is that these cells the number of these cells will actually scale to their correct size so that you all you get the same final outcome with
00:26:06
the different numbers as well the most amazing thing is that when you make these cells absolutely gigantic so that only one cell is big enough to make the whole Lumen what it will do is it will no longer cooperate with other cells one
00:26:19
cell will bend around itself making uh the exact same Lumen okay now the incredible thing about that is this is a different molecular mechanism this was cell to cell communication this is cytoskeletal bending
00:26:32
and so what happens is that this is a kind of a kind of top-down kind of top-down causation where in service of a large-scale anatomical spec meaning having the correct Lumen you can call up different molecular mechanisms to get
00:26:45
the job done so this again is sticking with this this theme of intelligence is the ability to uh handle novelty in terms of getting to where you're going from diverse starting positions uh with perturbations both external and internal
00:26:58
you know your own parts are getting are changing can you still do what you need to do and all of this is described in in some of these papers uh and then and then the final thing is that your your walk through morphe space doesn't even have
00:27:09
to be the same path so here for example here's a here's a frog frogs normally do not regenerate their legs as I'll tell you momentarily we have figured out a way to make them regenerate their legs and when they do so this is a pretty
00:27:22
good uh frog leg regeneration you can see here it's got the it's got the toes the toenails the webbing I mean everything's good but in fact the way it got here is not at all how frogs normally regenerate their limbs uh the
00:27:35
normally develop their limbs so normal frog limb morphogenesis is here you make these things and then you sort of make that you you kill off the cells in the middle to make the pad of the paddle this is this grows in a completely different way you've got the you've got
00:27:47
the middle this kind of middle stock here with a toenail and then the toes sort of come off to the side it looks much more like a plant so it takes a different path through morphe space but it ends up in the same place so all of
00:27:58
this all of this is is uh is is back to illustrating some of these amazing uh abilities of of these cells to um uh to to to navigate and so one very kind of uh very very canonical example of this
00:28:12
that we discovered a few years ago is is this so here's a tadpole here's the gut the brain the nostrils and the eyes here this tadpole needs to become a frog in order to for a tadpole face to be a frog face things have to move so the Jaws
00:28:24
have to move the eyes have to move forward everything has to move and it used to be thought that this process was hardwired because if you're a standard tadpole and you want to be a standard frog all you have to remember is which direction and by how much every piece of
00:28:36
the face moves what we did to and and we suspected that there was more intelligence to this process than that and so we did an experiment we we created so-called Picasso frogs and so these are tadpoles in which everything
00:28:49
is messed up the eyes on the side of the head the Jaws are off to the side and the nostrils are are too far back I mean everything is in the wrong position and we found that these animals still largely make pretty normal frogs because
00:29:01
all of these pieces will move in novel paths in fact sometimes they go too far and have to double back to give you a normal frog face so what the genetics gives you is not some hardwired system that always moves in the same way what
00:29:14
it specifies is a really interesting error minimization machine that however you start it off with obviously with some limits uh we'll try to minimize the error and get to the correct final shape
00:29:26
if we had a robotic swarm a collection of robots that was able to do this we would we would call this a prize-winning example of collective intelligence we don't have such a such a technology yet so uh so so we we started trying to
00:29:39
understand this process how how does all this work and so to the standard feed forward kind of open loop process of Developmental biology that that you that you would read about in in in class where there are genes they make proteins
00:29:52
there's there's some the proteins interact via some physics and chemistry and then there's this emergent outcome we add to this these feedback loops whereby this is actually a homeostatic system if that if that Anatomy is
00:30:05
disrupted in some way by injury by by um by injury by mutations by teratogens by parasites whatever then these feedback loops will kick in to try to minimize error the cells will do what they can to try to get back to the
00:30:18
correct shape it's a thing about your thermostat it's a basic homeostatic circuit now on the one hand this is a pretty pretty expected biologists know all about feedback loops um and and so on on the other hand there's there are
00:30:32
two kind of weird weird and unusual things here the first is that every homeostatic process has to have a set point so if you're going to try to get back to uh where you need to be you have to remember where the right
00:30:44
the where the right position is you have to store a set point we're used to thinking about scalars single numbers as set points so temperature pH things like that but in this case the set point is a some sort of a large scale uh geometry
00:30:57
it's a it's a descriptor of some kind of core screen descriptor of an anatomy so it's a complex data structure and in general you know biologists don't love to think about uh goal directed processes the idea is they're supposed
00:31:10
to be emergence and and kind of emergent complexity but this idea that things are working towards a goal the way that any navigational system fundamentally does is really not something that is is very
00:31:22
comfortable certainly for for molecular biology so how would something like this how would something like this work how could we have a navigating uh system that that can uh can have goals in anatomical space
00:31:34
and so here's where we start to think about bioelectricity because our best example of a a mechanism that allows us to navigate space towards goals is uh the brain and in the brain we know
00:31:47
roughly what the architecture looks like there are cells that communicate with other cells and networks these cells have ion channels so these are little proteins that help set a voltage to the cell and that voltage may or may not be
00:32:00
communicated to its neighbors through these little Gap Junctions these are like um like electrical synapses that allow electrical information to pass back and forth and so so that's the hardware of the brain and what that Hardware enables
00:32:12
is a kind of software that among other things navigate spaces so here's a zebrafish this is a movie uh taken by uh by by this group here that shows all the electrical activity in the living
00:32:23
zebrafish brain and so the commitment of Neuroscience is that if we understood how to decode this information we would be able to know what the cognitive content of this of this brain was so so the the memories
00:32:37
the preferences the goals whatever else the system was going to do we should be able to decode it so this is this is that cycle it's called a neural neural decoding right we should be able to understand what what all these patterns mean well it turns out
00:32:51
that uh this is not just for brains this is an extremely ancient system um all the cells in your body have ion channels most cells have Gap Junctions with each other and what we might be able to do is just like neuroscientists we might be able to extend this whole
00:33:04
scheme to ask what are your tissues thinking about at any point in time specifically to read the electrical so this is a this is a time lapse of an early frog embryo the colors are a fluorescent dye reporting voltage the
00:33:17
same way that we did here that was done here with these with these zebrafish uh and uh we might be able to decode this information to ask what are the Targets in anatomical amorphous space what is this thing going to build
00:33:30
and so there's this there's this amazing uh isomorphism between the story of the brain where you've got you've got the the hardware and the software and there's various experiences and various other ways that this software gets
00:33:42
modified um the hardware basically only just gets built by uh in in development and what what the brain does is it controls muscles to move your body in three-dimensional space now that's a
00:33:54
pretty cool trick where did it learn this trick well evolutionarily this is exactly the same as something that was going on long before brains appeared which is this exact same system but it
00:34:05
used to be for moving the body configuration in morphospace before there were nerves and muscles um cells needed to talk to each other and process information to get around in
00:34:16
morphe space and so uh We've argued in this in this paper that basically uh neural neural electricity is just an evolutionary pivot across spaces from morphe space to three-dimensional space
00:34:29
of a very a very ancient system for processing information long before brains arose so so in my group we then we we were thinking about this and we asked this question okay if that's if that's if that's uh how it works could
00:34:43
we make some tools to read and write this electrical information and to really understand how it's navigating that uh that that shape change space so we developed the some of the first tools to um uh read and write information into
00:34:56
non-brain tissues so of course these die so these are voltage sensitive fluorescent dyes that uh report all the electrical conversations that these cells are having with each other so here's a time lapse of these cells figuring out who's going to be left
00:35:10
right dorsal ventral so so all of this is is uh you're looking at these cells having those conversations um we do a lot of computational modeling to ask okay given where the ions are going how can we predict these voltages
00:35:22
I'm going to show you um two patterns here's a here's a time lapse it happens to be grayscale but it's the exact same idea uh here's the time lapse of a frog embryo putting its face together and what you will see in this this is one
00:35:34
frame out of that video what you will see is that this is this is something we call the electric face discovered by my colleague Danny Adams um when she was in my group doing this profiling she found that at here's
00:35:46
here's a particular frame from that uh from that video where you can see the the animal's right eye the mouth the uh the the the placards all of this is already demarcated uh before the genes start to come on and really pattern the
00:35:59
face this is a bioelectric pre-pattern of what the future face is going to look like and I'm showing you this one because it's so simple it literally looks like a face right there are other patterns that are much more complex that you can't really just you know sort of visually decode
00:36:12
but this one is very is very clear this is the memory of what a uh tadpole face is going to look like and if you move this uh if you if you disrupt this electrical pattern the gene expression will change and the organogenesis will change
00:36:25
um and uh we'll I'll show you these to show you examples of that momentarily so so this is a this is a uh an endogenous pattern that is required for normal development in fact human channelopathies of the human patients
00:36:39
that have craniofacial defects often have mutations and ion channels that that screw up this pattern now that's so that's a normal pattern here's a pathological pattern we've put in a human oncogene these cells will make a
00:36:50
tumor but even before the tumor becomes apparent you can already see this bioelectrical aberrant bioelectrical signature of these cells basically disconnecting from the rest of the tissue and going to their unicellular
00:37:01
ancient Behavior mode where the rest of the body is just environment as far as they're concerned so so so one one set of tools is to track these bioelectrical changes now here's another important set of tools to actually start to be able to uh to
00:37:14
write new information into it and the way we do that we don't use any applied electric Fields there are no waves there are no magnets there's no electromagnetic radiation what we do is we do exactly what neuroscientists do
00:37:25
which is we um modify uh the endogenous mechanisms by which cells establish electrical signals so either ion channels we can open them we can close them we can use light or drugs to do that
00:37:39
um and and the Gap Junctions the the electrical synapses so we can determine which cells talk to each other cells by opening and closing these little Gap Junctions it's very telling that all of the tools of Neuroscience work in other cell types basically the tools can't
00:37:51
tell the difference right this this distinction between neuroscience and and other cells in your body is completely artificial um and it's you know it's a consequence of humans trying to parse the world it's not um nature doesn't doesn't obey that distinction
00:38:04
so what can you do with this okay I've been talking about these this electrical information what what does it actually do well here's here's a few here are a few examples one thing you can do is you can take this this tadpole and I've shown you the
00:38:15
electric face and so we asked the question okay if now that we know that there's a particular voltage state that kickstarts eye development could we simply reproduce that somewhere else so we took an ion Channel RNA that we knew
00:38:27
was going to set the same voltage State we injected it into precursor cells that are going to give rise to the gut and sure enough those cells can form a complete eye they the eyes that are formed will have all the right layers uh
00:38:40
retina nerve you know optic nerve all that lens and uh and and and they can be formed anywhere they can be formed anywhere anywhere in the body as long as the cells get the right pattern that tells them what organ to make now notice
00:38:54
two very interesting things here one is that this instruction is highly modular we provided a very small piece of information just a voltage pattern we certainly did not give it all the information needed to specify an eye
00:39:06
eyes have many cell types um they're very complex there's no we have no idea how to micromanage the creation of an eye but what we provide I did was a kind of a subroutine call us a signal that says build an eye here and the cells do
00:39:17
that's the first thing the second amazing thing is that there are two levels of instruction here one is US instructing the cells build an eye but here what you can see is this is a lens sitting out in the flank of a of a of a
00:39:29
tadpole somewhere and the blue cells are the ones that carry this ION channel that we put in but there's not enough of them to make a good lens so what they've done is recruit their neighbors this is why at the very beginning I pitched this idea that frogalitals might have legs
00:39:43
that consist also of of of of frog cells why would frog cells build a leg when when they normally don't it's because cells have the ability to instruct each other about what to build and so these cells right here they they've been told to make an eye but there's not enough of
00:39:56
them and they know that and so they they recruit a bunch of their neighbors these Brown cells that we never touched they're not these cells are completely wild type and yet they're participating in making this ectopic lens out in the tail of a tackle so so there's this
00:40:09
amazing ability to to instruct and for the cells to then instruct each other now what else can we make well we can make in using the same methodology by manipulating these ion channels we can make ode assists which are balanced
00:40:22
organs like inner ear type of organs we can make complete beating Hearts so so ectopic Hearts we can make extra forebrain and then you can see you can ask if these animals are any smarter than the the normal tadpoles we can make
00:40:35
ectopic limbs so you can see here uh all kinds of extra extra legs that we can form and we can even make fins now this last one's kind of weird because tadpoles of course aren't supposed to have fins we'll get to that momentarily well what you can see is that by
00:40:48
changing the bioelectrical pattern using using a manipulation of ion channels you can tell this Collective where to go in morphe space meaning what types of organs they should be building you're not micromanaging the specifics each
00:41:00
step that they take they figure that out on their own in fact we have no idea how to how to do that for any of these complex organs but you can you can indicate regions about where it is that they should be going and this has has biomedical implications so frogs I told
00:41:14
John like salamanders do not regenerate their legs so if if a leg is amputated 45 days later there's nothing what we've done is come up with a a drug cocktail that targets ion channels that
00:41:25
um induces leg regeneration and so you can see here 45 days later there's already a nice a nice leg forming the leg is touch sensitive it's motile it's it's functional it has a pretty good eventually it has a pretty good uh final
00:41:37
a final pattern and so we're in the process of uh Translating that to mammals hopefully this will work I have to do a disclosure here because this is a commercial Venture now a morpheuticals Inc is a spin-off between myself and
00:41:50
Dave Kaplan where they the David's lab makes these bioreactors that uh provide that have an aqueous environment for the wound and then we provide the Ion channel and other types of payload in there to hopefully um get leg
00:42:03
regeneration in mammals so this is a clinically um kind of you know clinically relevant approach so I want to switch from this now to show you another kind of amazing example of um what bioelectricity does uh this is
00:42:16
these are planarium so so I've already told you they have incredible amounts of uh regenerative ability because of that they're actually Immortal there's no such thing as an old planarian they do not age so um these animals pretty much
00:42:29
go on forever uh and uh and and that's because of this incredible ability to continuously replenish and regenerate to keep their keep their pattern perfect so I want to show you I want to show you an example this is this is a planarian one head one
00:42:42
tail you cut off the head and the tail you got this middle fragment uh we check we check the the gene expression yep anterior genes are in the head no anterior genes in the tail that's fine and 100 of the time it makes this normal
00:42:55
Offspring so so one head one tail okay this it regenerates now here what I'm going to show you is that uh here here's this here's a planarian one head one tail uh again anterior genes where they're supposed to be but when I cut
00:43:08
this guy he makes a two-headed form and I just told you that this process was very reliable why in the world would a one-headed animal uh make a two-headed uh and this isn't Photoshop these are real you know real real live animals so so here are heads at both ends well it's
00:43:21
because in the meantime we found that there's an electrical circuit that actually helps this tissue remember how many heads they're supposed to be and what you can do in the meantime is take this one-headed animal and you can look at the electrical pattern ah
00:43:34
the electrical pattern says one one head one tail and we can alter that electrical pattern to say no you're going to be you a proper worm should have two heads now you can see it's kind of a mess the technology is still being worked out but but it's very clear we
00:43:46
can we can we can impose this pattern this is two heads and when you cut that animal boom that's what the cells make now this is really critical to understand this bioelectric pattern is not a map of this two-headed guy this
00:43:58
bioelectrical pattern is a map of this one-headed animal into which we incepted a false memory of what planaria are supposed to look like and they sit there this memory sits there late into doing nothing until you injure them once you
00:44:11
injure this animal that is when they will the cells will pay attention to this pattern and they will go ahead and make a two-headed animal so if you're wondering where counter factual um where the ability to consider counterfactuals came from this kind of mental time travel where you can
00:44:24
consider uh things that have memories and have predictions about things that are not happening right now this is perhaps the uh the sort of uh basal precursor of that ability in morphe space this this animal has a
00:44:38
representation in morphospace of where it will go if it gets injured at a future time not what's happening right now that's not this is not where I am right now this is where I'm going to go if I get injured now why do I keep calling this thing a memory I keep
00:44:50
calling it a memory because if you take these two-headed animals and simply re-cut them again in plain water no more manipulations of any kind uh they will continue to to rebuild two-headed animals in forever as far as
00:45:03
we can tell the the question of how many heads a planarian has is not strictly determined by The genome because we didn't touch the genome The genome is wild type there's nothing there's no genomic editing here the the that
00:45:15
information is sitting in an electric circuit the default the genetically said default is one head but it's easily changed and in fact we can now well now as I say it's easy it took us 10 years to you know to figure it out but but in
00:45:27
fact it's it's quite readily done and in fact you can uh you can set it back you can set it back to being one-headed by by again targeting the information in that electrical circuit um the first time uh I showed I showed these data somebody stood up at a
00:45:40
conference and said well that's impossible those animals can't exist so I always bring this video so everybody can see what uh what the second and third generation of these guys look like uh so so this this has all the properties of memory it's long long-term
00:45:53
stable it's rewritable it's got a little bit of labability so it's rewritable it's got latency or conditional recall so you can have a single body can have memories of what a planarian is supposed to look like that aren't true right now
00:46:05
and it has discrete possible outcomes so one of the things that we're doing now is is really trying to integrate models of the physiological State space of of this animal that where where the
00:46:17
circuit the electrical circuit has has a state space where there are retractors that correspond to one heads two heads no heads and so on and merge that with models of connectionist artificial
00:46:28
neural network type of processes where where we already know how to build networks that for example do pattern completion and they have a memory of what the pattern should be and if part of that part of that pattern is missing they can reproduce it and so on this is
00:46:41
all this is all part of one problem and and the key is tying it all together in uh with these quantitative models that show how this electrical circuit navigates morphos space now what I've shown you is that you can tweak that
00:46:53
electrical circuit to find an attractor that has two heads instead of one heads but what else is there what else does this space actually have let's explore that a little bit well the first thing you find is that well it contains head shapes of other species so what you can
00:47:06
do is you can take this planarian cut off the head uh perturb the electrical circuit for about 48 hours then let them let it go and when it finally settles down into the correct attractor it doesn't always find the right one so
00:47:20
some percent of the time it goes back to normal and makes this nice triangular head from uh the Dorado cephala but sometimes it'll make a round head like this s Mediterranean sometimes it'll make a flathead like this pflena in fact
00:47:32
this is a stochastic process the frequency of these uh is proportional to The evolutionary distance between the real um of the real species that are being mimicked here and again no changes to
00:47:45
the genome there's nothing genetically wrong with these animals but you can you can find these other attractors and it's not just for head shape it's the shape of the brain the distribution of the stem cells become just like these other species and it's about they're about 100
00:47:58
to 150 million years distant okay so so you can find those attractors with the exact same uh genetics uh in fact you can find shapes that uh don't that that don't correspond to any real planarian
00:48:11
as far as we know you can make these crazy uh spiky forms you can make a different uh symmetry type so this kind of cylindrical radial symmetry instead of a flat bilateral symmetry you can make combinations uh where where there's
00:48:23
a flatworm with a big tube growing out into the third dimension and again nothing wrong with the hardware there's not no no no mutations another one with the genome it's exploring the it's bioelectrically exploring the space of
00:48:34
where it can be and so the idea is what we're doing now is is building these uh kind of full stack models that that try to integrate all the way from the molecular information about what channels do you have to work with so the
00:48:47
hardware all the way up through the physiology of okay well what does this mean for the voltage States across the tissue to well what does that mean for the identity of different organs along the primary axis to what kind of
00:49:00
algorithm is this whole thing is this whole thing implementing such that we could actually understand the algorithm and intervene right so so make make changes and for this we have we have all kinds of computational platforms such as
00:49:13
such as this which allows you to load up electrical circuits into cells and then um simulate assimilate tissue and organ level Dynamics to understand what the large-scale computations are hmm
00:49:26
and so so just as a simple example again of of why this is why this is useful and what the what the Practical implications of this are we use this the strategy to find a a technique for repairing birth
00:49:38
defects so here's a tadpole here's the forebrain midbrain and hindbrain if it's exposed to nicotine or various other teratogens you can see there's massive defects of the the brain is basically gone um the forebrain is basically gone the
00:49:51
midbrain and hybrid are damaged and so what we did is we built this this bioelectrical model that tries to explain how does the brain know what its size and shape should be and in cases where for example here this is even
00:50:03
worse than a teratogen we mutated a key neurogenesis Gene this is a gene called Notch so this Gene is mutated you can see the forebrainage is gone the midbrain and hindbrain are a bubble these animals have no Behavior they just lay there doing nothing it's a very
00:50:15
strong defect what we found what we did was we asked that model um Which ion channels would we have to open and close to get back to the correct bioelectrical memory of what a what a correct brain should be basically how do I find my way back in morpha
00:50:28
space to what a correct brain should be and the model made a prediction we tested that prediction it happens to be a a an already human approved um set of drugs that can do this and sure enough
00:50:40
when we did it there there you go uh the the uh their brain structure comes back their brain um gene expression comes back and their IQ comes back so if you actually test them for their learning rates they're indistinguishable from controls even
00:50:54
though they've been exposed to teratogens or they have this really nasty genetic defect some problems and not all problems but some problems can be fixed at the level of software sometimes when your Hardware isn't quite right you can make up for it in in
00:51:06
software by having a better navigation policy through morphospace so you're not dead and you're not automatically dead in the water if your Hardware is a little wonky so what that means is uh we ought to be able to create these
00:51:18
pipelines to uh to and we've already we've already started um to help design electroceuticals that is you you have some particular problem and you can ask the simulation what type of Channel
00:51:30
meaning what drug can I use to get back to my correct bioelectrical pattern fundamentally this is really the the question of how do we exploit this electrical interface that these cells are exposing to us right just like
00:51:43
neurons do and people using um electrodes and optogenetics and everything else to control brains using this electrical interface we can do exactly the same thing to help guide their navigation through morphospace it's a new way of to think about getting
00:51:57
making drugs for uh various disorders of of of development of injury and cancer and so on okay so so I want to just just say a couple of things uh before I wrap up here I want to say a couple of things
00:52:09
about uh the scaling of cognition in this whole in this whole business and then show you some some novel synthetic living machines so the first thing I want to point out is that we've been I I started out by saying that thinking
00:52:23
about fixed uh standard animals is is kind of limiting and that's because the border between self and world is is flexible it can change so so again here
00:52:36
is a and and what what unifies all all of these different kinds of weird intelligences on solving problems in morphe space and in physiological space and so on is there ability to pursue goals meaning to pursue the right region
00:52:50
of that morphospace so this kind of creature a single cell can have pretty humble little tiny goals right so so on the scale of some number of microns it can it can pursue uh State
00:53:03
physiological States so metabolic States and so on but these cells can cooperate and when they Co-op when they cooperate they can pursue very large goals so building a limb is a huge goal not no individual cell knows what a limb is or
00:53:15
how it can count fingers or anything like that but the collective sure can the collective will absolutely pursue this particular region of morphus space despite all kinds of perturbations and when does it stop it stops when it when
00:53:27
it reaches that area so when it gets to where it's going and the limb is correct that's when it stops so there's a scaling of goals here now that happens during evolution and as you as you as you'll see in a minute it also can can happen right in
00:53:39
front of your eyes but the reverse process we're quite familiar with and that's cancer so these here are glioblastoma cells human glioblastoma cells and culture cells can defect from this uh kind of um from this kind of
00:53:52
situation once they electrically isolate from their neighbors then as far as they're concerned they're back to they're back to this scenario the the rest of the body is just external environment to them their their goals are little tiny goals what are the goals
00:54:04
of single cell systems well for every cell wants to become two cells and it wants to go wherever life is good and that's metastasis and so that process that's shrinking and growing in fact uh these cancer cells are not any more selfish than your normal body cells it's
00:54:18
just that the self is smaller right a lot of a lot of work in um game theory of of cancer that talks about selfishness and lack of cooperation they're not any more selfish it's just that the the self towards which they work is just little little tiny itself
00:54:30
is very small whereas here the electrical network is able to bind and if people want I have I have all sorts of stories about what exactly happens when they bind together but I'll skip that for now for reasons of time when
00:54:43
when individual cells bind into networks those networks have a greater ability to to store memory to have anticipation of future events to perceive spatial uh spatial kinds of signals in their
00:54:55
environment and so the scale grows and the uh the biomedical implication of this is that if this were If This Were true you should be able to reverse cancer by artificially reconnecting
00:55:10
cells to their neighbors and in fact that's what we've that's what we've done so so when you have this this nasty oncogene you can see that in red it's still here you can prevent the tumor from happening by putting in an ion channel that forces these cells to
00:55:22
remain in the right bioelectrical State despite the fact that the oncogene is trying to disassociate them and if you do that the physiology trumps the genetics it will it will these cells will remain even though the oncogene is strongly expressed the cells will remain
00:55:35
uh of making nice muscle and skin and whatever else it's supposed to make so so this this idea of looking at the scaling of the self and the scaling of these goals I was real uh practical uh implications and in in my in my model
00:55:49
one of the things that um it allows the framework to do is to compare directly very diverse intelligences so any intelligence be it evolved designed uh some sort of hybrid some sort of alien
00:56:02
thing maybe you saw pure software it doesn't matter any agent has a set that can pursue goals you can simply plot the size of those goals so so how big are they in space and time right you might have you might have a tick that only
00:56:15
cares about local butyrate concentration and that's all that it's ever going to care about and so it has a tiny little um cognitive light cone that allows it to pursue these tiny little goals anymore or you might have a dog which has a bigger cognitive light cone and
00:56:27
has some pretty good memory going backwards has a little bit of anticipation potential forwards it's never going to care about what happens Two Towns over three months from now it's impossible that its goal space is simply not that big right and then you
00:56:39
can have humans which uniquely perhaps have uh have a cognitive light cone that's bigger than their own lifespan so the the human can comprehend and pursue goals that are that are guaranteed not achievable in your lifetime and that for
00:56:52
all I know that may drive some interesting psychological pressures that that humans face but this idea that that we are all made of a collective of Agents each of these agents is solving problems in its own space and within
00:57:04
that space there are differently scaled goals some some very very you know very modest and some some massive you know some humans are working for world peace and and you know these kinds of very complex very large long-term things so
00:57:16
the idea is that in this multi-scale system what happens is that higher levels bend the option space for their subunits so so in uh if if you've got a
00:57:28
morphogenetic space that that uh that that has attracted for different head shapes that's because those those bioelectrical States distort the the space of gene expression for their cells such that all the cells need to do is go
00:57:41
down their concentration gradients the way they normally do they don't have to know where they're going but in fact they end up at a very particular morphological outcome because because the higher level has already distorted
00:57:53
that space and so there are all kinds of interesting work to be done looking at mathematical formalisms from relativity Theory and and from some other disciplines to look at how and how that works and in fact all that's happening
00:58:05
here is the scaling up of goals so once you have a single cell that can do this cycle this this test measure ACT test compare act kind of cycle let's say for example for keeping pH once you start
00:58:18
connecting cells into electrical networks now everything scales up their goals are bigger their memories are bigger the the ability to act are bigger this this is what evolution is is doing it's constantly scaling up these these
00:58:30
homeostatic units and it's actually uh pivoting them through different spaces so very simple kinds of organisms uh all they could do is Traverse metabolic space to keep alive but eventually that works up to
00:58:43
physiological spaces then gene expression becomes a thing and so they can work in transcriptional spaces then multi uh the complex Anatomy so so morphe space and eventually when brains and muscles show up you're in behavioral space and who knows what other spaces
00:58:56
there are linguistically and there may be many on this so just to close off in the last few minutes um I want to I want to show you an example of novelty uh so I've been I've been talking about intelligence and the idea that
00:59:08
um these uh all of these different uh different systems are able to handle novel conditions I want to show you an extreme example of that so this is uh this is work that uh Joshua bongard's lab and and we are doing in uh in in
00:59:20
this new Institute and we wanted to ask a simple question uh well a set of questions one is how much how much real-time plasticity really is there I mean we've seen that you can get you can get pretty far away from the genomic
00:59:32
default but what what else can these cells do and in fact when we look at development normal development to what extent is it is it fooling us to what extent are we basically lulled into thinking that because oh acorns make oak
00:59:45
trees and frogs eggs make make frogs to what extent are we really fooled into thinking that that's basically all that they can do and so we started asking about you know what what might the default competencies of some of these
00:59:57
cells and tissues be okay and so so what we did was was this some simple experiment I'm just going to uh pause oops I'm going to pause this here what we did was a simple experiment where um and this all the biology was done by
01:00:10
Doug Blackiston um in my group in the in the computer science you're about to see was done by uh Sam kriegman um what we did was we took some skin cells off of a frog embryo and we put those skin cells by themselves in a dish so we didn't add
01:00:23
anything there are no Nano materials there are no genetic circuits being added there are no weird trans genes or chemicals what we do is subtract what do we subtract we subtract all this other stuff we subtract all their normal neighbors that are normally uh
01:00:35
instructing these cells as to what to do when we give them a chance to reboot their multicellularity now there's all sorts of things they could do so so they could just go and die they could spread out and get away from each other they could form a flat two a two dimensional
01:00:49
monolayer the way that cell culture does there are many things that that could happen and so so here's what happens we um we dissociate these cells we put them in this little little depression well overnight they sort of come together like this and then they they form this
01:01:02
little round ball this the the flashes you're seeing is calcium signaling so it's a calcium sensitive diet it um you they they start to sort of signal to each other and what they form is something we call xenobots now why
01:01:14
xenobots because xenopus lavis is the name of the frog and uh and it's a biorobotics platform so the way it's the way they're they're swimming is they have little hairs on their surface these hairs called cilia are normally used to
01:01:26
redistribute mucus on them across the body of the Frog here what they've done is they've sort of they've learned to row and so and so the cells the the silly are moving and this thing is propelling itself they can go in circles they can go straight back
01:01:39
and forth like this uh here you can see a bunch of them in their various uh this is just tracking data so so these two are interacting these are sitting there doing nothing this one's going on kind of a longer Journey
01:01:51
um here here's one navigating a maze this is still water there's no water movement in here so it moves along it takes the corn without bumping into the opposite wall and then at some point for
01:02:03
reasons that nobody nobody can predict yet it decides to turn around and go back where it came from so it again goes along does not need to bump into anything to know that it could take a corner here and then at this point
01:02:15
turns around turns around and goes and goes back okay now um one of the things you can see with this calcium signaling is that they're very active this looks very brain-like this looks very similar to that zebrafish kind of signaling that that I
01:02:28
showed you before except there's no neurons here this is just skin everything that you just saw was 100 skin cells there's no there's no brain there's no nervous system and so not only uh you know so so who knows we're still experimenting with like what what
01:02:41
might they be saying to each other when they're when they're doing this but but one thing you could you can see they do a couple of interesting things one thing they can do is regenerate so here's here's one that was basically cut in half and now think about think about the
01:02:52
force look at that hinge as it as it folds up think about the the force that it takes to to clamp something shut from 180 degree position but it basically folds up right to be um back to its uh
01:03:04
to its xenobot shape and one of the things we started thinking about is how can we how can we predict what this kind of system is going to do I mean nobody knew what what they were going to do so so Josh bongard started modeling their
01:03:17
their behavior in this kind of simulation the simulation can actually tell you what the behavior of different shaped Bots is going to be and then you can go ahead and micro sculpt them or do various things you can see the patterns that this is just stuck Carmine powder
01:03:30
on the bottom of the dish and these guys sweep it along as they go so you can see all the cool patterns that they make as they're navigating but but here's something here's something interesting if you if you'd um simulate them in an environment with a
01:03:42
bunch of loose bricks what you see is that they tend to they tend to collect them into little piles and so what's what might be the significance of that well it turns out the real Bots do exactly that and if you provide them so these are
01:03:55
xenobots provided with a bunch of loose skin cells so these white dots are just loose skin cells hanging around and what you'll see is that these the the xenobots will run around they will collect either together or individually
01:04:07
they'll make little piles and because this is in a gentle material right these cells are not passive they once the cells are collected into a pile what do they make they make xenobots so this is kinematic self-replication these Bots
01:04:20
make copies of themselves from the loose material in the dish we've made it impossible for them to replicate the way that normally frogs reproduce so they can't do that but within 48 hours they show us that
01:04:32
there's another way to do this that as far as we know no other animal uses it's basically Von Neumann's dream which is uh machines that will go out and build copies of themselves from parts that they find laying around that's what's going on here these xenobots build the
01:04:45
next generation of xenobots out of these loose cells when those mature guess what they do they run around do exactly the same thing you get the Next Generation and then the next Generation after that now there's no strong heredity here yet
01:04:56
um as far as we know but uh and and the reason it works it is really critical the reason it works is the exact same reason that we were able to make xenobots in the first place because these cells are not passive Legos they
01:05:09
are in a gentle material that once they get together in a particular group they know what to do they they have an innate preference to to make Bots that's that's why these Bots are able to make newbots so uh here's the uh here's kind of the
01:05:22
amazing thing about it is that um same genome so if you were to sequence the xenobot all you would ever see is the zenipa slavis genome you'd have no idea that it was any different that same genome can do at least two different uh
01:05:35
life histories it can do this so is that so normal normal xenopus um uh uh stages in the normal tadpole Behavior or it can do this so here's an early xenobot here's a xenobot at about two months
01:05:47
it's it has a developmental sequence it's turning into something what is that I have no idea what that is there's never been there's never been any xenobots before so it's some kind of Novel uh novel a developmental stage and eventually they they have these these
01:05:59
kinds of behaviors so we're still working out we don't know how much cognition they actually have can they live can they learn can they uh they have preferences be what do they react to We don't know we're still figuring that out but the amazing thing
01:06:11
about it is that uh from for most animals the way they navigate both physical and anatomical space you when you ask you know why do they have a certain number of legs a certain color certain behaviors so the answer is always well because for millions of
01:06:25
years the ancestors were selected for this or that and everybody else died off and now this is what you have well there's never been any xenobots and there's never been selection to be a good Zen about all of this is created on the Fly this is this is what they do
01:06:36
they create a coherent organism and by the way if you ask what do skin cells normally want to do you might think that well what they normally want to do is to be this like passive two-dimensional layer on the outside of an animal to keep out the bacteria that isn't what
01:06:50
they want to do what that's what they're bullied into doing by the instructive interactions with the rest of the cells if you don't have those cells that this is what they're actually this is their default this is actually their default Behavior so what evolution does is what
01:07:01
bioengineers can and will be doing is behavior shaping it's figuring out signals to get your a gentle material not passive not a blank slate but you're a gentle material to do whatever you want to do I'm just going to um close up
01:07:13
here with with a couple of thoughts this is uh this is just a kind of a an overview of this of this framework that that we've been working on is kind of Continuum of agency across very diverse implementations
01:07:26
the idea of of scaling of the boundaries of goals for any particular system as a way to compare radically different types of intelligences in different spaces bioelectricity is a particular example not the only one I'm sure there are
01:07:39
other ways to tell this story but bioelectricity is a really great way to look at what is the uh what is the cognitive medium of a collective intelligence and it seems to be electricity in many cases um and there are all kinds of
01:07:51
implications for for evolution including asking questions about where do these goals come from if it's not just selection where in fact do they come from and so just to finish up I want to point out that one of the consequences
01:08:04
of the fact that biology can solve problems on the Fly and that it's incredibly interoperable is that pretty much any combination of evolved material meaning cells or tissues or genes design
01:08:16
material so smart Nano materials all kinds of constructs and and software any combination of these is some viable creature and so when Darwin said endless forms most beautiful impressive being
01:08:29
impressed with a variety of biology that he was seeing around that's just all of that everything on Earth is a tiny corner of this incredibly huge option space which we're already beginning to explore so there are already cyborgs you
01:08:41
know humans with various implants both sensory and um and effectors people can run wheelchairs with their minds and have I'll have all kinds of sensors attached to them uh High brots which are brains driving Vehicles instead of
01:08:53
biological bodies um where people started a long time ago making making chimeras so so mules right horses and donkeys make mules so so you can start to mix genetic material after that people would make grafts and plants
01:09:05
and mix at the cellular level every combination we are going to be living in a world where we are surrounded by agents that don't look like anything from familiar on the phylogenetic treatment and this means that we have
01:09:19
massive implications for systems of Ethics because the old traditional framework where you look at something and you ask where did it come from would you come from a factory or was it was it natural and what does it look like does
01:09:31
it look like a fish an ape or a human or you know or or something else those kinds of touchstones are going to be completely useless in uh deciding how to relate to these beings that are going to
01:09:43
be around us in terms of what do we owe them what do uh what can we expect of them all of these things we we can we're not going to be able to rely on familiar categories of where did it come from and how much like a like a human does it you
01:09:55
know does it look that's going to be gone so those things will not survive the next few decades so we are going to need to develop novel Frameworks because life is so good at navigating these spaces that all of these things are going to be viable so I'll just close by
01:10:07
saying that um all of this is is discussed in great detail in various papers anybody just please email me and I'll send you everything um I want to thank the the various intelligent beings that contributed to all the work I showed you so all of our
01:10:19
postdocs PhD students the model systems that we work with they do all the heavy lifting all the all the slime molds and the planarian the tapples and everything else and of course our funders and again uh two disclosures so morpheuticals is a
01:10:33
company around limb regeneration and fauna systems is a company for xenobots and that's it uh thank you so much thank you for listening
End of transcript